CurledWake: full divergence form for delta_u diffusion

[zanig]

Replaces the Laplacian-only diffusion term nu_T * (d2u/dy2 + d2u/dz2)
with the full divergence form d/dy(nu_T du/dy) + d/dz(nu_T du/dz)
in CurledWakeWindfield._step. Also drops the now-unused second_der import.

I believe this is a leftover from previous versions where the turbulence models were a lot simpler (const $\nu_T$ or neglible gradients), since The Δk transport equation in CurledTurbulenceModel_kl.compute_dkdx uses the full divergence form already for diffusion.
A nice side effect of this fix is that with scipy_rk45, the full divergence form requires only roughly a thrid of integration steps compared to the current Laplacian-only version on example case 08, so this should also improve run time.

Impact: Taking example case 08 with the full divergence form vs. the current Laplacian-only implementation, I see the following differences in Cp:

• Turbine 2: ~1.5%
• Turbine 3: ~6%

You're making a pull request to a branch (probably main) of MITWindFarm. Please ensure you have done the following.

• Request a review from other Howland Lab members who use MITWindFarm.
• Add at least a sentence on your change to the documentation (probably the quickstart guide).
  ---> no new functionality so I assume this is not needed?
• Add tests for your new functionality.
  ---> more of a "physical" fix so explicitly testing is difficult
• Make sure the tests pass and the documentation notebook still runs.
• Get approval from the folks you requested a review from.

If you want more details on best practices, please see the following guide on the Howland Lab Google Drive.

Happy merging!